基于深度学习的MIMO-OFDM盲源分离算法OA

This paper addresses blind separation of OFDM signals in multiple-input multiple-output(MIMO)systems under non-cooperative communication scenarios.To overcome the insufficient separation accuracy of traditional algorithms under phase ambiguity and emission angle variations,an end-to-end multichannel blind separation network based on time-domain beamforming is proposed.The network employs a two-stage architecture that integrates inter-channel and channel-specific features,and utilizes DPRNN to estimate beamforming filters for OFDM signal separation.To enhance robustness to variations in emission angles,a transform-average-concatenate(TAC)module is integrated between DPRNN blocks for information alignment and feature fusion.In addition,a customized loss function is designed to effectively resolve the phase inversion problem.Simulation results demonstrate that under 4QAM with a signal-to-noise ratio(SNR)of 14 dB,the proposed method reduces the bit error rate(BER)below 10-3,thus significantly outperforming traditional blind source separation algorithms and verifying the effectiveness of the network.Furthermore,its low computational complexity and end-to-end structure provide practical feasibility for future engineering applications.